Re: [PATCH 1/1] iommu/vt-d: Fix missed device TLB cache tag

[Baolu Lu]

On 6/20/24 11:04 AM, Tian, Kevin wrote:
> > From: Baolu Lu<baolu.lu@xxxxxxxxxxxxxxx>
> > Sent: Thursday, June 20, 2024 8:50 AM
> >
> > On 6/20/24 12:46 AM, Jason Gunthorpe wrote:
> > > On Wed, Jun 19, 2024 at 09:53:45AM +0800, Lu Baolu wrote:
> > > > When a domain is attached to a device, the required cache tags are
> > > > assigned to the domain so that the related caches could be flushed
> > > > whenever it is needed. The device TLB cache tag is created selectively
> > > > by checking the ats_enabled field of the device's iommu data. This
> > > > creates an ordered dependency between attach and ATS enabling paths.
> > > >
> > > > The device TLB cache tag will not be created if device's ATS is enabled
> > > > after the domain attachment. This causes some devices, for example
> > > > intel_vpu, to malfunction.
> > > What? How is this even possible?
> > >
> > > ATS is controlled exclusively by the iommu driver, how can it be
> > > changed without the driver knowing??
> > Yes. ATS is currently controlled exclusively by the iommu driver. The
> > intel iommu driver enables PCI/ATS on the probe path after the default
> > domain is attached. That means when the default domain is attached to
> > the device, the ats_supported is set, but ats_enabled is cleared. So the
> > cache tag for the device TLB won't be created.
> I don't quite get why this is specific to the probe path and the default
> domain.

The issue is with the domain attaching device path, not specific to the
probe or default domain.

> dmar_domain_attach_device()
> {
> 	cache_tag_assign_domain();
> 	//setup pasid entry for pt/1st/2nd
> 	iommu_enable_pci_caps();
> }
>
> seems that for all domain attaches above is coded in a wrong order
> as ats is enabled after the cache tag is assigned.

Yes, exactly. But simply changing the order isn't future-proof,
considering ATS control will eventually be moved out of iommu drivers.

> why is it considered
> to affect only some devices e.g. intel_vpu?

This bug was discovered during testing of the intel_vpu device and
affects devices other than the intel_vpu. The commit message is a bit
confusing. :-)

Best regards,
baolu